Spatial scale effects on associations between built environment and cognitive function: Multi-Ethnic Study of Atherosclerosis

Abstract

Built environments have the potential to favorably support cognitive function. Despite growing work on this topic, most of the work has ignored variation in the spatial scale of the effect. The issue with spatial scale effects is that the size and shape of the areal unit within which built environment characteristics are measured naturally influence the built environment exposure metric and thus the estimated associations with health. We used spatial distributed lag modeling (DLM) to estimate how associations between built environment exposures (walkable destinations [WD], social destinations [SD]) and change in cognition varied across distance of these destinations from participants’ residences. Cognition was assessed as maintained/improved processing speed (PS) and global cognition (GC). Person-level data from Exam 5 (2010–2012) and Exam 6 (2016–2018) of the Multi-Ethnic Study of Atherosclerosis was used (N=1380, mean age 67). Built environment data were derived from the National Establishment Time Series. Higher availability of walkable and social destinations at closer distance from participants’ residence was associated with maintained/improved PS. The adjusted associations between maintained/improved PS and destinations waned with increasing distance from the residence; associations were evident until approximately 1.9-km for WD and 1.5-km for SD. Associations were most apparent for participants living in areas with high population density. We found little evidence for associations between change in GC and built environment at any distance. These results highlight the importance of identifying appropriate spatial scale to understand the mechanisms for built environment-cognition associations.

1. Introduction

Built environment has drawn growing interest due to its potential to promote healthy behaviors and positive health outcomes (Garin et al., 2014; Renalds et al., 2010; Cummins et al., 2007). The built environment can be broadly defined as environmental spaces that have been planned/modified/adapted for human utilization. This includes a wide range of things: from buildings and sidewalks to retail establishments and public spaces like parks (Renalds et al., 2010; Saelens & Handy, 2008). Based on the socio-ecological model, built environmental factors could function as a mechanism influencing people’s cognition (Besser et al., 2017). Built environment features such as walkable and social destinations could directly impact cognition by exposing individuals to various levels of cognitive load and multisensory stimulation which are closely related to attention and memory performance (Finlay et al., 2022; Cassarino & Setti, 2015). Built environment destinations could also indirectly affect cognition through mobility behaviors (i.e. walking, cycling) and lifestyle behaviors such as physical and social activities (Cerin, 2019; Cassarino & Setti, 2015). There has not been a lot of work in this area and findings have been somewhat mixed (Chen et al., 2022; Besser et al., 2017). Better cognitive health has been associated with greater access to park space (Cherrie et al., 2018; Besser et al., 2021a), better access to public transit (Clarke et al., 2015), and greater street connectivity (Watts et al., 2015). Less work has been done on access to destinations needed for daily living such as food stores, pharmacies, and banks. One recent study found that living in neighborhoods that had many commercial and public destinations nearby was associated with maintained/improved cognitive health (after adjusting for numerous individual variables as well as contextual factors such as neighborhood SES (Besser et al., 2021a); whereas, two other studies found no significant associations between cognitive function and neighborhood availability of two different types of destinations: health care facilities and recreational centers (Luo et al., 2019; Clarke et al., 2012).

Inconsistent results between neighborhood built environment and cognitive function may be attributable to the operationalization of neighborhood measures. Specifically, the selection of which aspects of the built environment to measure (e.g., social destinations, greenness, walkability) and the spatial scales used to define built environment exposures may impact whether an association is detected. Identifying a relevant spatial scale to measure built environment exposures and their associations with health outcomes is challenging because area-based spatial measures are subject to the Modifiable Areal Unit Problem—the size and shape of the areal unit will influence the built environment exposures and their associations with health (Fotheringham & Wong, 1991). Indeed, empirical evidence illustrates that associations between built environment and health behaviors vary depending on the spatial scales defined and delineated (Chen et al., 2022; Gonzales-Inca et al., 2022; Labib et al., 2020; Frank et al., 2017; James et al., 2014; Kwan, 2012a; Spielman & Yoo, 2009). For instance, a study found that associations between healthy food accessibility and grocery store location choice were different across buffers around the residence with varying sizes of 1-km, 2-km, and 3-km, respectively (Li & Kim, 2020). Previous research on built environment-cognition associations has mainly focused on predefined fixed spatial scales around residence within which built environment characteristics were measured. The fixed spatial scales were traditionally defined and delineated either using buffers around residence or using administrative boundaries. For example, most studies defined the spatial scale for characterizing built environment in buffers around residence with varying sizes of 0.25-km (Paul et al., 2020), 0.5-km (Zhu et al., 2020; de Keijzer et al., 2018), 0.8- km (Besser et al., 2019; Koohsari et al., 2019; Watts et al., 2015), 1-km (Tani et al., 2019), 1.5-km (Cherrie et al., 2018), 1.6-km (Astell-Burt et al., 2020). Some others measured built environment in administrative boundaries (e.g. census tract, census block) (Brown et al., 2018; Bastos et al., 2015). A few studies included two spatial scales like 0.8-km and 1.6-km to compare the associations between built environment and cognition across multiple spatial scales (Besser et al., 2021a; Cherrie et al., 2019; Wu et al., 2017). However, there is often little theoretical or empirical evidence to support the spatial scale, and using fit criteria such as R-squared to select the buffer size performs poorly (Spielman & Yoo, 2009).

Second, a pre-specified spatial scale does not account for spatial autocorrelation between built environment exposures inside the spatial unit and those in areas that are adjacent to the predefined spatial unit. This effect, referred to as the residential effect fallacy, results in biased estimates of the health effects of built environment exposures that are assessed for a pre-specified spatial scale with a fixed radius around a residence if the correlation between built environment inside and outside the boundary of the defined spatial scale are ignored (Chaix et al., 2017). Further, a pre-specified spatial scale may not represent an individual’s perceived neighborhood, also known as the Uncertain Geographic Context Problem (Kwan, 2012b). For example, 1.6-km circular and 1.6-km network buffer, represents less than one-half of the neighborhood boundaries identified by residents of those neighborhoods (Smith et al., 2010). In addition, pre-defined spatial scales may not be appropriate for different types of built environment features. Specifically, the spatial scale of an association with cognition may be different for various elements of the built environment if the underlying mechanism or impact on human behavior varies. For instance, empirical evidence indicates that the relevant spatial scales for associations of walkable destinations and frequent social destinations with walking for transportation purposes are different (Li et al., 2022a). Similarly, the spatial scale of walkable environments’ impacts on cognition may work through increased transport walking with a larger buffer compared to the spatial scale of social destinations’ impacts on cognition, which may impact cognition through alternative pathways such as social connections with close neighbors.

A major methodological challenge for developing interventions is identifying the most relevant spatial scales within which built environment features are most influential for older adults and their cognitive function. It is critical to employ an easy-to-implement method to mitigate misspecification bias associated with pre-defined spatial scales while estimating the built environment and health associations (Hu et al., 2023; Lovasi et al., 2011; Martin et al., 2014). Recent studies employed a distributed lag modeling approach to circumvent pre-defining spatial scales and estimate associations of built environment features with health behaviors like transport walking (Li et al., 2022a) or health outcomes like body mass index (BMI) (Baek et al., 2016; Baek et al., 2017) across a set of consecutive ring-shaped network buffer areas. The distributed lag modeling employs a set of “distributed lag exposures”, instead of a single built environment exposure within a predefined spatial scale. Importantly, these studies revealed the distance decay phenomenon in the built environment—health associations across increasing distances from residence. Their results enabled us to identify the threshold point where the associations between built environment and health become negligible, facilitating identification of the appropriate spatial scales for built environment—health pairs. The findings in these studies highlight that the relevant spatial scales vary depending on different pairs of built environment feature—health behavior/outcome. These models have been shown to have both conceptual advantages and yield better inferences (e.g., less bias) compared to traditional models using buffer-based approaches (Baek 2016). Therefore, it is necessary to use methods that enable us to investigate and identify the relevant spatial scale for other pairs of built environment feature and health outcomes, and thus obtain more accurate built environment-health associations.

In addition to methods-oriented research showing different scales for different exposure-outcome pairs, it is possible that cognitive function has a different spatial scale than other health outcomes. First, lower cognitive function is a function of aging which is not independent of other aging processes, including limits to physical function and mobility (Ng et al., 2018). These factors may result in smaller relevant spatial scales, or more condensed neighborhoods. Furthermore, the changes in cognitive function with age might shift the relevant spatial scale by limiting comfort with new environments or navigating farther destinations. Consequently, implementing novel tools to determine what spatial scale is relevant to estimate associations between built environment and change of cognitive function may inform place-based interventions. Despite these issues, to our knowledge, no study investigates the spatial scale effects on built environment features and change of cognition.

Further, based on the socio-ecological model, an individual’s sociodemographic characteristics may affect cognitive function (Finlay et al., 2022; Cerin, 2019; Besser et al., 2017), which may impact the relevant spatial scale for an individual. Empirical evidence has illustrated that age (Gard et al., 2014), sex (MacAulay et al., 2020), race/ethnicity (Clarke et al., 2012), education (Darwish et al., 2018), and household income (Shen et al., 2018) have been related with cognitive function. These factors have similarly been associated with life space, a measure of older adult mobility that shows how engaged an individual is in their environment and how far from home they may travel (Eronen et al., 2016; Phillips et al., 2015). Besides, racial-ethnic differences in underlying factors associated with cognition, such as financial situation, quality of education, and cost-related reduced medication use (Gordon et al., 2020). Historical racist policies like redlining have been associated with present-day neighborhood environment and health disparities (Swope et al., 2022). Related, individuals’ health condition like diabetes and suspected clinical depression are associated with cognition (Rexroth et al., 2013), and with life space metrics (Polku et al., 2015; Brown et al., 2009; Baker et al., 2003). Additionally, empirical evidence illustrated that living in more urbanized areas is related with better cognitive performance (Cassarino et al., 2018). Further, neighborhood socioeconomic disadvantage may be associated with cognitive function (Guo et al., 2019) and spatial scale (Caldas et al., 2020; Allman et al., 2006).

To address this gap, our study aimed to employ distributed lag models (DLM) to examine spatial scales of associations between built environment and cognitive change based on data from the Multi-Ethnic Study of Atherosclerosis (MESA). We aimed to estimate cognitive change associated with built environment destinations across varying distances from residence. We focused on two built environment features identified previously as impacting cognitive function in MESA: walking destinations and a subcategory of walking destinations, frequent social destinations (Besser et al 2021b). Neighborhoods with more walkable destinations and social frequent destinations may encourage people to walk to visit these destinations (Li et al., 2022a), which could help maintain late-life cognition (Ng et al., 2018). We executed this work using a distributed lag modeling (DLM) approach. The DLM approach used in this study could reveal distance-decay phenomenon in the built environment-cognitive change associations and identify the most relevant spatial scales for the two types of built environment destinations on cognitive change, respectively.

2. Methods

2.1. Data and study sample

This analysis uses person-level data from the Multi-Ethnic Study of Atherosclerosis (MESA) cohort, which recruited 6,814 adults (39% Non-Hispanic White, 22% Hispanic, 28% African American, and 12% Chinese American) aged 45–84 recruited from six study sites (Los Angeles, CA; Chicago, IL; Saint Paul, MN; New York, NY; Baltimore, MD; Forsyth County, NC) starting in 2000. Participants resided in urbanized areas. Cognitive assessments were done in two later exams: Exam 5 which occurred during 2010–2012, and Exam 6 which occurred during 2016–2018. MESA addresses were geocoded using TeleAtlas (Tele Atlas North America, Inc., Lebanon, New Hampshire). Road network data used for network analysis (described below) were obtained from ESRI Business Analyst 2016. Out of the 6814 enrolled at baseline, 3150 participated in Exam 5 and Exam 6. For sample inclusion, first, we retained participants who completed a cognitive assessment, and who had no ICD-9 diagnoses for dementia determined from hospitalization or death records (Fujiyoshi et al., 2016), and no dementia medication use. This selection yielded a sample of 1622 participants. Second, we excluded 98 participants who had missing built environment variables. We further excluded 144 participants who had missing sociodemographic information. The final sample consisted of 1380 participants.

Neighborhood built environment data were derived from National Establishment Time Series (NETS) longitudinal database from Walls & Associates (Walls & Associates, Denver, CO). These data are not a survey or statistical sample; instead, they represent a continual annual census of American business, government, and non-profits. Methods on their collection can be found elsewhere (Walls, 2015). We used NETS data spanning 1990–2014, selecting years that aligned with MESA Exam 5 (2010–2012). The NETS was originally obtained and processed through the MESA Neighborhoods study (Hirsch et al., 2022) and Retail Environments for Cardiovascular Disease (RECVD) project using methodology described elsewhere (Hirsch et al., 2021).

2.2. Variables

2.2.1. Health outcomes

We focused on two measures of cognitive function. The validated (Jian et al., 2021) Cognitive Abilities Screening Instrument (CASI) (Teng et al., 1994) which is a global measure of cognitive function assessing attention, concentration, orientation, abstraction, judgment, verbal fluency, language, short-term and long-term memory, and visual construction (range: 0–100). The validated (González-Blanch et al., 2010) Digit Symbol Coding (DSC) was a subtest of the Wechsler Adult Intelligence Scale-III (Wechsler, 1997) that measured processing speed (range: 0–133). Higher scores indicate better cognitive function. We first calculated the change scores of the raw CASI and DSC between Exam 5 and Exam 6, respectively. For use in our analysis, we further dichotomized the change scores of CASI and DSC into two categories: 0 indicated decline in cognition and 1 indicated maintained/improved cognition. Category 1 combined maintenance and improvement because cognitive decline frequently accompanies aging thus maintaining cognition is a goal for healthy aging (Bishop et al., 2010); additionally, a very small number of participants improved cognition which may be partly attributed to the practice effect (Machulda et al., 2013), thus cannot be a standalone category.

2.2.2. Exposures: Count of built environment destinations

We grouped destinations into two categories determined to be important to health conditions (Li et al. 2022a; Li et al., 2022b). The first category was “walkable destinations”, which included destinations that were common for daily life (food stores, restaurants, drug stores and pharmacies, department stores, post offices, banks/credit unions, libraries, beauty shops and barbers, social/entertainment destinations, museums, schools, etc.). The second category was a smaller subset of the first: “frequent social destinations” that facilitated social interaction and promoted social engagement (e.g., beauty shop/barber). Methods for classifying these destinations are described elsewhere (Hirsch et al., 2021), and a detailed list of destinations is shown in Supplement Table 1.

2.2.3. Covariates

All covariates at individual/household level including sex, gender, race/ethnicity, and education came from Exam 1. Other covariates at individual/household level came from Exam 5. Covariates included factors that could affect where an individual resides and/or utilization of built environment/walking around one’s neighborhood, and could also affect cognitive function. These were identified through literature reviews (Chen et al., 2022) and variables used by others when studying place effects on cognition (Besser et al., 2021b; Ng et al., 2018). We included demographics (age, sex, race/ethnicity), socio-economic indicators (education, annual family income, employment status), health conditions (clinically measured body mass index, hypertension, diabetes; and self-reported depressive symptoms (CES-D), emphysema/chronic obstructive pulmonary disease, arthritis, smoking, and self-rated health), and genotype apolipoprotein E (APOE, the strongest genetic risk factor for late onset Alzheimer’s disease and related dementias (Gharbi-Meliani et al., 2021)). Additionally, we also adjusted for neighborhood-level socioeconomic status proxies by percent of household income with above median income in 2010–2012 [>= $50,000] at census tract level and population density at census tract level (U.S. Census, https://www.census.gov).

2.3. Distributed lag modeling

Distributed lag models (DLMs) are widely used in economics (Almon, 1965) and air pollution studies (Dominici et al., 2004; Pope III et al., 1991; Welty et al., 2009). Most recently, DLMs have been used to estimate associations between spatially-lagged built environment exposures and transport walking (Li et al., 2022a) as well as body mass index (Baek et al., 2017; Baek et al., 2016).

In order to estimate how the association between built environment destinations and cognition vary with network distance (distance measured along road networks) between participants and destinations, the DLM uses a set of “distributed lag exposures”, instead of a single count of destinations within a network buffer of pre-determined radius. The DLM accounts for spatial autocorrelation since it estimates the effects of built environment exposures in a set of consecutive ring-shaped network areas that covered both the close and faraway areas from residence. The distributed lag covariates are the count of built environment destinations for participant i measured within a ring-shaped network buffer area with inner and outer radii of $r_{l-1}$ and $r_l$, respectively, and are denoted as $X_i\left(r_{l-1};r_l\right)$ for $l=\mathrm{1,2},\dots ,N$. We set the width of the ring as $r_l$-$r_{l-1}$= 100-meter, which is a typical width for city blocks in U.S. cities. In a hypothetical city with 100 × 100 meter blocks, the distributed lag covariates would thus represent the number of destinations available within consecutive groups of blocks incrementally further away from the residence. We set the largest ring-shaped network buffer area to have an outer radius of 5-km (3.11 miles), as in a prior study very few individuals are willing to walk farther than this (Li et al., 2022a). Hence, the inner and outer radii of the first ring-shaped network area were 0 and 0.1-km; the inner and outer radii of the 50th ring-shaped network area were 4.9-km and 5-km. Distributed lagged exposures were constructed in this way for walkable destinations and social destinations, respectively.

2.3.1. DLMs for MESA cognitive data

The DLMs for MESA data were then:

$$\text{log}\left(\frac{p_i}{1-p_i}\right)={\beta }_{0i}+{\sum }_{l=1}^{50}\beta \left(r_{l-1};r_l\right)X_i\left(r_{l-1};r_l\right)+\gamma Z_i+{\epsilon }_i.$$ (1)

Where $p_i$ denotes the probability that the change score of a cognitive measure between Exam 5 and Exam 6 for participant $i$ is equal to 1 (i.e., the probability of maintaining/improving cognition); the predictor $X_i\left(r_{l-1};r_l\right)$ denotes count of built environment destinations in the ring-shaped network buffer area with inner and outer radii of $r_{l-1}$ and $r_l$, respectively for participant $i$ at Exam 5; $\beta \left(r_{l-1};r_l\right)$ denotes the corresponding distributed lag coefficient, quantifying the association between the count of destinations in said ring-shape area and cognitive change between Exam 5 and Exam 6; and ${\mathit{Z}}_{\mathit{i}}$ denotes a set of covariates for participant $i$ at Exam 5.

Additionally, we also wanted to know if distance to walking and social destinations differs when the overall number of services/retails is low vs. high. Built environment and cognition may only be evident for residents living in areas where there is sufficient density of services/retail that they can habitually rely on (Cassarino et al., 2018). Population density is positively correlated with density of services/retail (Li et al., 2022b; Besser et al., 2021b), thus can serve as a rough proxy for density of services/retail. We classified population density into below median vs. median or above (>=5770 per square mile or 15000 per sq km) and then stratified the DLMs described above.

2.3.2. Results presentation and interpretation

First, we used figures to display DLMs results on estimated distributed lag coefficients quantifying associations between cognitive changes and built environment destinations up to 5-km from residence adjusting for covariates. Each figure demonstrates a series of distributed lag coefficients in ring-shaped network buffer areas with ranges from 0–0.1km up to 4.9–5km. Second, we presented aggregated values of these coefficients, which represented associations within ring-shaped network area associations that have wider width (i.e. 0.5 km). The aggregated association in a given ring-shaped network area with inner and outer radii of $r_m$ and $r_k$ were derived by summing the coefficients within the range. For instance, $\overline{\beta \left(r_m;r_k\right)}=\sum _{l=m}^k\beta \left(r_{l-1};r_l\right)$. We presented a series of aggregated ring-shaped network area associations with ranges of 0 – 0.5km, 0.5 – 1km, 1 – 1.5km, 1.5 – 2km, 2 – 2.5km, 2.5 – 3km, 3 – 3.5km, 3.5 – 4km, 4 – 4.5km, 4.5 – 5km. Additionally, we calculated aggregated buffer area association $\overline{\beta \left(0;r_k\right)}$ for a given network buffer with a radius of $r_k$ from the DL beta coefficients to facilitate comparison with prior studies. The aggregated buffer area association $\overline{\beta \left(0;r_k\right)}$ was derived by summing the coefficients within the buffer with a radius of $r_k$, i.e., $\overline{\beta \left(0;r_k\right)}={\sum }_{l=1}^k\beta \left(r_{l-1};r_l\right)$. Ten aggregated buffer area associations in respective network buffer areas with radii of 0.5km, 1km, 1.5km, 2km, 2.5km, 3km, 3.5km, 4km, 4.5km, and 5km were presented. We back-transformed each aggregated association $\overline{\beta \left(r_m;r_k\right)}$ and $\overline{\beta \left(0;r_k\right)}$ by $\text{Exp}\left(10*\overline{\beta \left(r_m;r_k\right)}\right)$ and $\text{Exp}\left(10*\overline{\beta \left(0;r_k\right)}\right)$ to aid interpretation. Note that we multiplied the aggregated buffer-based associations $\overline{\beta \left(r_m;r_k\right)}$ and $\overline{\beta \left(0;r_k\right)}$ by 10 for better readability and interpretation. The transformed associations represent the average odds ratio associated with 10 additional built environment destinations in the corresponding buffer-based areas (with buffer ranges of $r_m$-$r_k$, or 0-$r_k$, respectively) to maintained/improved cognition.

3. Results

3.1. Descriptive statistics

The average age of participants was about 67-year-old (standard deviation [SD] = 8, min-max: 53 – 91) with about one-half women (Table 1). The median CASI score for global cognition was 91.2 (Q1–Q3: 85.5–95.4), and the median DSC score for processing speed was 55 (Q1–Q3: 45–67) at Exam 5. The distribution for CASI at Exam 5 was what would be expected based on their age (and generally higher education), and the DSC score at Exam 5 was relatively low, which may be attributable to the age and diversity of this cohort (Fitzpatrick et al., 2015). About 54% and 35% participants had maintained/improved global cognition (CASI score) and processing speed (DSC score), respectively after approximately 6 years of follow-up. Approximately 75% had some college degree or above, and 54% of participants had an annual household income equal to/greater than $50,000. Around 35% participants were obese, 31% participants had hypertension, 13% participants had depressive symptoms, 2% participants had emphysema/chronic obstructive pulmonary disease, 17% participants had diabetes, and 17% participants had arthritis flareup in last 2 weeks. About 27% of participants had ≥1 apolipoprotein E (APOE) genotype ε4 allele. About 53% of participants lived in neighborhoods (census tracts) with median household income greater than $50,000. The mean population density of the residential neighborhoods (census tracts) was 13,400 people per square mile (SD = 18,400 people per square mile).

Table 1.

Descriptive statistics for sample characteristics (N=1380)

Outcome variables	Exam 5	Exam 6
	Median (Q1–Q3)	Median (Q1–Q3)
Cognitive Abilities Screening Instrument (CASI): (possible range 0–100)	91.2 (85.5 – 95.4)	92 (87 – 95.4)
Digit Symbol Coding (DSC): (possible range: 0–133)	55 (45 – 67)	51 (38 – 63)
Cognitive change between Exam 5 and Exam 6	%
Maintained/Improved CASI between Exam 5 and Exam 6	53.77
Maintained/Improved DSC between Exam 5 and Exam 6	35.07
Covariates at Exam 5
Age: Mean (SD)	66.9 (8.23)
% female	52
Race/ethnicity	%
White	42.6
Chinese	13
African American	25.8
Hispanic	18.6
Education
High school degree or under	25.3
Some college, no bachelor’s degree	30
Bachelor’s degree or higher	44.8
Annual household income
<$25,000	21.1
$25,000–$49,999	25.1
$50,000–$74,999	19.3
>=$75,000	34.5
Percent of participants who were employed	25.6
Obese (body mass index >= 30kg/m2)	34.5
% hypertension	31.16
Depressive symptoms (% of ceds score >= 16)	12.6
% Emphysema/chronic obstructive pulmonary disease	2.1
% with diabetes	17
% with Arthritis flareup in last 2 weeks	17
% currently smoking	6.6
Self-rated health
Better	62.1
Same	34.2
Worse	3.7
Apolipoprotein E (APOE) genotype ε4 allele	26.4
% of participants living in higher income neighborhoods (> $50,000): Mean (SD)	53 (17.7)
Neighborhood population density per square mile: Mean (SD)	13400 (18400)

Table 2 shows the descriptive statistics for built environment characteristics in aggregated ring-shaped network buffer areas and aggregated buffer areas with varying ranges. There were a high number of destinations around participants’ addresses and the numbers increased in ring-shaped network areas as they got farther away from residence. For instance, the median number of walkable destinations in the ring-shaped network area with inner and outer radii as 0-km and 0.5-km was 9 (Q1–Q3: 2–29), and the median number in the ring-shaped network area with inner and outer radii as 1.5 and 2km was 82 (Q1–Q3: 30, 201.2). The median values for frequent social destinations within ring-shaped network areas with varying buffer ranges show a similar pattern. As expected, the median values for walkable destinations and social destinations in ring-shaped network areas that were closer from residence were smaller. These values increased in ring-shaped network areas that were farther away from residence.

Table 2.

Descriptive statistics for built environment characteristics

	Ring-shaped area: ${\mathit{R}}_{\mathit{m}}-{\mathit{R}}_{\mathit{k}}$	Median (Q1–Q3)	Buffer range: ${\mathit{R}}_{\mathit{m}}-{\mathit{R}}_{\mathit{k}}$	Median (Q1–Q3)
Walkable destinations	0–0.5km	9 (2–29)	0–0.5km	9 (2–29)
	0.5–1km	37 (9–104.2)	0–1km	46 (12–134)
	1–1.5km	62.5 (20–160)	0–1.5km	111.5 (32.75–298)
	1.5–2km	82 (30–201.2)	0–2km	193 (66–494)
	2–2.5km	106.5 (45.75–260.5)	0–2.5km	304 (117.8–755.2)
	2.5–3km	144 (55–300.2)	0–3km	471 (182.8–1024.2)
	3–3.5km	173.5 (70–353.5)	0–3.5km	655 (262.8–1338)
	3.5–4km	201 (81–428.5)	0–4km	835 (342–1741)
	4–4.5km	222.5 (91–466.2)	0–4.5km	1059.5 (437.8–2191.8)
	4.5–5km	244 (101.5–488)	0–5km	1263 (531–2534)
Frequent social destinations	0–0.5km	1 (0–4)	0–0.5km	1 (0–4)
	0.5–1km	4 (1–13)	0–1km	5 (1–17)
	1–1.5km	7 (2–22)	0–1.5km	14 (4–37)
	1.5–2km	9 (3–27)	0–2km	23 (8–66)
	2–2.5km	11 (5–36)	0–2.5km	36 (13–103)
	2.5–3km	16 (6–41.25)	0–3km	51 (19–142)
	3–3.5km	18 (7–51)	0–3.5km	73 (28–191.2)
	3.5–4km	20 (9–57)	0–4km	89.5 (38–254)
	4–4.5km	25 (9–61)	0–4.5km	114 (47–315.2)
	4.5–5km	29 (10–66)	0–5km	138 (58–369)

3.2. Spatial scale effects in the associations between cognitive change and built environment

Figure 1 displays adjusted associations between cognitive change and built environment for the varying distances from the participant’s residence. The associations between cognitive change measure of DSC (processing speed) and walkable destinations varied across distances (Figure 1). The associations were stronger at closer distances from participants’ residence. The associations became negligible at the distance of 1.9-km. The associations between cognitive change measure of DSC and frequent social destinations also decayed with increasing distances (Figure 1). The associations were stronger at closer distances from participants’ residence and vanished at the distance of 1.5-km. Meanwhile, the associations became negative at farther distances beyond 1.5-km, indicating that the odds of having maintained/improved processing speed were lower at farther distances. Note that the association between change cognitive measure of DSC and frequent social destinations was stronger than the association between change cognitive measure of DSC and walkable destinations at the same distance. In addition, we found that the cognitive change measure of CASI (global cognition) was not associated with walkable destinations or frequent social destinations regardless of varying distances (Supplement Figure 1).

Figure 1.

Distributed lag modeling results for cognitive change (processing speed) and built environment destinations in ring-shaped network areas up to 5-km from residence.

Note: y-axis represents log odds. Each model adjusted age, gender, race/ethnicity, education, annual household income, employment status, BMI, arthritis flareup in the last 2 weeks, self-rated health, APOE ε4 carriers, depression, diabetes, emphysema/chronic obstructive pulmonary disease, hypertension, smoking, neighborhood population density, and neighborhood socioeconomics. N = 1380.

Table 3 displays estimated average ring-shaped network area association between cognitive change and built environment destinations in the ten aggregated ring-shaped network areas and in the ten network buffer areas. Participants living in areas with 10 additional walkable destinations in the ring-shaped network area with inner and outer radii of 0 and 0.5km had 3% increased odds of maintained/improved processing speed (measured by DSC). The magnitude of the odds became smaller in the ring-shaped network areas that were a greater distance from residence. Regarding frequent social destinations, 10 additional frequent social destinations in the ring-shaped network area with inner and outer radii of 0–0.5km were associated with 20% increased odds of maintained/improved cognition. The average associations also became smaller in the ring-shaped network areas with inner and outer radii of 0.5–1km and 1–1.5km. In addition, the average buffer-area associations between processing speed and walkable destinations and frequent social destinations existed up to 3-km and 2.5-km, respectively. Meanwhile, for global cognition (measured by CASI), there was high uncertainty in the estimates and no clear pattern and none of the buffers were statistically significantly associated with walkable destinations or frequent social destinations in either ring-shaped areas or buffer areas with varying ranges (See Supplement Table 2).

Table 3.

Estimated associations between change scores of processing speed and built environment destinations within specific ring-shaped areas and buffer areas

N = 1380	Ring-shaped areas (Rm-Rk)		Buffer areas (0-Rk)
	Buffer ranges (km): Rm-Rk	Exp(10*beta(Rm;Rk), 95% C.I.	Buffer ranges (km): 0-Rk	Exp(10*beta(0;Rk), 95% C.I.
Walkable destinations	0–0.5km	1.03 (1, 1.05)	0–0.5km	1.03 (1, 1.05)
	0.5–1km	1.02 (1, 1.04)	0–1km	1.05 (1.01, 1.1)
	1–1.5km	1.02 (1, 1.03)	0–1.5km	1.07 (1.01, 1.14)
	1.5–2km	1.01 (1, 1.03)	0–2km	1.09 (1.01, 1.17)
	2–2.5km	1.01 (0.99, 1.02)	0–2.5km	1.09 (1.01, 1.19)
	2.5–3km	1 (0.99, 1.01)	0–3km	1.1 (1, 1.2)
	3–3.5km	1 (0.99, 1)	0–3.5km	1.09 (0.99, 1.2)
	3.5–4km	0.99 (0.99, 1)	0–4km	1.08 (0.98, 1.19)
	4–4.5km	0.99 (0.98, 1)	0–4.5km	1.07 (0.96, 1.19)
	4.5–5km	0.99 (0.98, 1)	0–5km	1.05 (0.94, 1.18)
Frequent social destinations	0–0.5km	1.2 (1.02, 1.42)	0–0.5km	1.2 (1.02, 1.42)
	0.5–1km	1.16 (1.01, 1.34)	0–1km	1.4 (1.03, 1.9)
	1–1.5km	1.12 (1, 1.25)	0–1.5km	1.57 (1.03, 2.38)
	1.5–2km	1.08 (0.99, 1.18)	0–2km	1.7 (1.03, 2.81)
	2–2.5km	1.05 (0.99, 1.11)	0–2.5km	1.78 (1.02, 3.11)
	2.5–3km	1.01 (0.98, 1.04)	0–3km	1.8 (0.99, 3.24)
	3–3.5km	0.98 (0.96, 0.99)	0–3.5km	1.76 (0.96, 3.23)
	3.5–4km	0.94 (0.91, 0.98)	0–4km	1.66 (0.87, 3.15)
	4–4.5km	0.91 (0.86, 0.97)	0–4.5km	1.51 (0.75, 3.06)
	4.5–5km	0.91 (0.85, 0.97)	0–5km	1.37 (0.63, 2.97)

Note: Each model adjusted age, gender, race/ethnicity, education, annual household income, employment status, BMI, arthritis flareup in the last 2 weeks, self-rated health, APOE ε4 carriers, depression, diabetes, emphysema/chronic obstructive pulmonary disease, hypertension, smoking, neighborhood population density, and neighborhood socioeconomics.

3.3. Sensitivity analysis results

Supplement Figure 2 and Supplement Figure 3 show adjusted associations between cognitive change and built environment varying across distance by population density. The associations of processing speed (measured by DSC) with walkable destinations and frequent social destinations across varying distances differed by population density. In higher density areas (population density above median 5770 per square mile), having maintained/improved processing speed was associated with both walkable destinations and social destinations at closer distances up to 2.9-km from residence. Meanwhile, in areas with lower population density, the association was not evident for either walkable or frequent social destinations and there was no evidence of effect modification for the CASI outcome.

4. Discussion

This study employed a distributed lag modeling approach to investigate the associations between cognitive change and built environment destinations across varying distances. The DLM approach enabled us to visualize the decay phenomenon in associations between cognitive change and built environment destinations across varying distances and identified the distance at which the associations became negligible. The positive associations between maintained/improved processing speed and walkable destinations decayed with the increasing distances from residence. Walkable destinations far away from a person’s residence may become less relevant for cognitive function as these destinations were beyond the person’s life space. The decay phenomenon also held for the associations between maintained/improved processing speed and frequent social destinations. The association coefficients were positive for both built environment categories at closer distances, indicating that higher availability of walkable destinations and frequent social destinations near residential locations were associated with maintained/improved processing speed. In addition, the DLM helped to distinguish the thresholds where the associations became negligible comparing between different built environment types. Our findings illustrated that the association became negligible at approximately 1.9-km for walkable destinations and 1.5-km for frequent social destinations, respectively.

The positive associations between change of processing speed and built environment destinations were stronger in closer distances from residence and weaker in farther distance. This finding supported the hypothesis that built environment exposures within closer distances from residence were more relevant for cognition. Both walkable destinations and frequent social destinations within closer distance from residences were associated with maintained/improved processing speed. One potential reason may be that higher number of built environment destinations could encourage a variety of errands and social practice and higher level of transport walking (Li et al., 2022a), which subsequently could lower cardiovascular risk and help maintain or improve processing speed (Ng et al., 2018; Baker et al., 2010). Our finding aligned with the results from a study in the same sample which indicating that walkable destinations in 1-mile (~1.6 km) buffer at Exam 5 were associated with maintained/improved processing speed, but not global cognition (Besser et al., 2021b).

Further, the spatial scales for walkable destinations and frequent social destinations on change of processing speed were approximately at 1.9-km, and 1.5-km, respectively. This result aligned with findings in previous literature indicating that the thresholds where the impacts of built environments and transport walking varied by type of built environment feature evaluated (Li et al., 2022a). One potential explanation is that the threshold distances to destinations where people are willing to visit shifts for different types of built environment destinations (Gunn et al., 2017; Yang and Diez-Roux, 2012; Moudon et al., 2006).

Besides, within the same spatial scale, the magnitude of association with cognitive processing speed was stronger for frequent social destinations than walkable destinations even though social is a subset of walkable. Walking destinations may be too global and broad, capturing destinations that include both those with associations to cognitive function and those with little to no impact. Additionally, walkable destinations likely promoted transport walking and physical activity which impacted cognition through fairly distal biological pathways (Cerin, 2019). In contrast, social destinations could lead to more social activity which can closely track with brain activity and processing (Paiva et al., 2023; Zhou et al., 2020). Overall, this finding further supports performing analyses that investigate various built environment types or measures and clearly operationalizing pathways between neighborhoods and cognitive function (Chen et al., 2022; Besser et al., 2017).

In addition, in our sample the change of global cognition (measured by CASI) was not associated with built environment destinations at Exam 5, regardless of varying distances. One potential explanation was that most global cognition measures were insensitive but had good specificity for cognitive impairment. This result was consistent with the prior MESA study illustrating no association between change of CASI and walkable destinations and social destinations at Exam 5 in both 0.5-mile and 1-mile, respectively (Besser et al., 2021b).

The association between maintained/improved processing speed and built environment destinations across varying distances was modified by population density. Associations between maintained/improved processing speed and both walkable destinations and frequent social destinations existed within closer distances up to 2.9-km from residence at denser areas with population density higher than 5770 people per square mile, but not at any distance in lower population density areas. A potential explanation for the divergence by population density may be that the availability of walkable destinations and frequent social destinations is higher in denser areas than that in sparser areas (Li et al, 2022b). The differences were also consistent with the idea that accessing to cognitive stimulating resources may differ between more and less urbanized areas (Wörn et al., 2017). Likewise, the threshold distance for associations between maintained/improved processing speed and frequent social destinations was 1.5-km in the full sample and 2.9-km in the stratified sample in highly populated areas. The different threshold distances for the same pair of cognitive change and built environment destinations by population density may indicate that the relevant spatial scales for each type of built environment measure may also depend on the density and spatial distribution of built environment destinations. Given that the MESA sample spans six different US cities with varying histories, development patterns, and densities, the modification we observed may represent city-level differences in behavior or norms around interacting with features of one’s neighborhood. More research is needed to investigate the relationships between cognitive change and different types of built environment destinations in various levels of population density across even more locations.

There are some implications of our DLM results about the spatial scale effects across varying distances. First, the DLM results reveal how associations between changes of cognitive functions and built environment destinations vary by distance from residence. Our DLM results allow us to identify the spatial scale within which built environment exposures are more relevant to cognitive functions, which could inform future research to select more appropriate spatial scales to characterize built environment exposures and cognition. Further, the DLM approach could allow us to target relevant spatial scales for specific built environment categories in specific geographic contexts. Ultimately, if replicated in future research and other samples, our findings could provide guidance on the appropriate scale for public health place-based interventions to promote healthy aging. For example, if distribution of destinations for cognitive enrichment (e.g. libraries, social clubs, etc.) is associated with slower cognitive decline, this could encourage land use policies that zone these uses in areas anticipated to experience increases in older adult populations. Similarly, measures of walkable social destinations could be incorporated into tools to aid older adults when making a decision where to live, fostering awareness and choices based on the importance of social activities to cognitive aging.

4.1. Strengths and limitations

Strengths of this study include that it is the first study to estimate how the associations between changes of cognitive functions and built environment destinations vary across distances using DLM. Second, we identified relevant spatial scales for walkable destinations and frequent social destinations to understand differences by type of environmental exposure or pathway to cognition. Importantly, we examined outcomes longitudinally, a significant improvement over cross-sectional research in ensuring appropriate temporal order of neighborhood exposures in relation to changes in cognition. However, it also has some limitations. First, due to data availability, we did not examine longitudinal change in built environment exposures and their associations with longitudinal change in cognition. Future research could extend this study by investigating the longitudinal associations between change of cognition and change of built environment exposures across distances using DLM. Second, we did not consider how cognitive function at baseline (exam 5) acting in the associations, which may produce potential bias in magnitude and direction from the total and direct causal effects (Tennant et al., 2022). Third, we only examined two types of built environment destinations and we did not take into account infrastructure for walkable destinations. We also did not adjust for other built environment variables like park access and air pollution which may be relevant to cognition (Cerin, 2019). Future research may want to investigate other built environment features or destinations with relevance for cognition in older adults. Fourth, although the built environment exposures in areas up to 5-km from residence may overlap with activity space, we did not account for built environment exposures around other daily activity locations (i.e. workplaces or volunteer locations) or activity paths beyond residential neighborhoods. Future research could incorporate built environment exposures within broader activity space into the DLM approach to enrich our understanding on cognitive change—built environment associations. In addition, future studies would benefit from more follow-ups and a more comprehensive battery of cognitive tests covering multiple domains. Prior work determined that individuals excluded due to missing cognitive data at two time points were more likely to be Chinese or Hispanic, have lower educational attainment and lower income, and live in neighborhoods with lower SES (Besser et al., 2023; Besser et al., 2021b). If these characteristics are associated with greater decline in cognitive health as well as residing in neighborhoods with fewer walkable and social destinations, then selection bias would have occurred and could have biased the estimate toward or away from the null. Finally, our results were derived based on a sample of older adults from relatively urban areas (i.e. all fall into metropolitan statistical areas), which may not be generalizable to other age groups (like children, adolescents, or younger adults) or to other geographic contexts (like rural areas).

5. Conclusions

The relevant spatial scales for a pair of cognitive change and built environment destinations vary depending on different types of built environment and different geographic contexts. The DLM approach allowed us to investigate how associations between cognitive changes and built environment destinations vary across varying distances. Our DLM results inform future research by providing information important for selecting appropriate spatial scales within which built environment exposures may be relevant for cognition. Thus, ultimately, this work could inform decisions by policymakers and urban planners about the placement of amenities to help maintain or improve cognitive function. These results highlight future research needed on developing methods to estimate associations between health outcomes and various built environment features across a variety of contexts, which could enhance our understanding on the spatial mechanisms in the built environment-health associations.